Multi-level Weighted Additive Spanners - arXiv
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Multi-level Weighted Additive Spanners∗
Reyan Ahmed
University of Arizona, Tucson, United States
abureyanahmed@email.arizona.edu
Greg Bodwin
University of Michigan, Ann Arbor, United States
bodwin@umich.edu
Faryad Darabi Sahneh
University of Arizona, Tucson, United States
faryad@email.arizona.edu
Keaton Hamm
University of Texas at Arlington, Arlington, United States
arXiv:2102.05831v3 [cs.DM] 29 Mar 2021
keaton.hamm@uta.edu
Stephen Kobourov
University of Arizona, Tucson, United States
kobourov@cs.arizona.edu
Richard Spence
University of Arizona, Tucson, United States
rcspence@email.arizona.edu
Abstract
Given a graph G = (V, E), a subgraph H is an additive +β spanner if distH (u, v) ≤ distG (u, v) + β
for all u, v ∈ V . A pairwise spanner is a spanner for which the above inequality only must hold for
specific pairs P ⊆ V × V given on input, and when the pairs have the structure P = S × S for
some subset S ⊆ V , it is specifically called a subsetwise spanner. Spanners in unweighted graphs
have been studied extensively in the literature, but have only recently been generalized to weighted
graphs.
In this paper, we consider a multi-level version of the subsetwise spanner in weighted graphs,
where the vertices in S possess varying level, priority, or quality of service (QoS) requirements, and
the goal is to compute a nested sequence of spanners with the minimum total number of edges. We
first generalize the +2 subsetwise spanner of [Pettie 2008, Cygan et al., 2013] to the weighted setting.
We experimentally measure the performance of this and several other algorithms for weighted additive
spanners, both in terms of runtime and sparsity of output spanner, when applied at each level of the
multi-level problem. Spanner sparsity is compared to the sparsest possible spanner satisfying the
given error budget, obtained using an integer programming formulation of the problem. We run our
experiments with respect to input graphs generated by several different random graph generators:
Erdős–Rényi, Watts–Strogatz, Barabási–Albert, and random geometric models. By analyzing our
experimental results we developed a new technique of changing an initialization parameter value
that provides better performance in practice.
2012 ACM Subject Classification Theory of computation → Design and analysis of algorithms
Keywords and phrases multi-level, graph spanner, approximation algorithms
Digital Object Identifier 10.4230/LIPIcs...
Supplement Material All algorithms, implementations, the ILP solver, experimental data and
analysis are available on Github at https://github.com/abureyanahmed/multi_level_weighted_
additive_spanners.
∗
This paper has been accepted in the 19th Symposium on Experimental Algorithms, SEA 2021.
© Reyan Ahmed, Greg Bodwin, Faryad Darabi Sahneh, Keaton Hamm, Stephen Kobourov, and
Richard Spence;
licensed under Creative Commons License CC-BY
Leibniz International Proceedings in Informatics
Schloss Dagstuhl – Leibniz-Zentrum für Informatik, Dagstuhl Publishing, GermanyXX:2 Multi-level Weighted Additive Spanners
Acknowledgements The research for this paper was partially supported by NSF grants CCF-1740858,
CCF1712119, and DMS-1839274.R. Ahmed, et al. XX:1
1 Introduction
This paper studies spanners of undirected input graphs with edge weights. Given an input
graph, a spanner is a sparse subgraph with approximately the same distance metric as the
original graph. Spanners are used as a primitive for many algorithmic tasks involving the
analysis of distances or shortest paths in enormous input graphs; it is often advantageous to
first replace the graph with a spanner, which can be analyzed much more quickly and stored
in much smaller space, at the price of a small amount of error. See the recent survey [5] for
more details on these applications.
Spanners were first studied in the setting of multiplicative error, where for an input graph
G and an error (“stretch”) parameter k, the spanner H must satisfy distH (s, t) ≤ k ·distG (s, t)
for all vertices s, t. This setting was quickly resolved in a seminal paper by Althöfer, Das,
Dobkin, Joseph, and Soares [8], where the authors proved that for all positive integers k, all
n-vertex graphs have spanners on O(n1+1/k ) edges with stretch 2k − 1, and that this tradeoff
is best possible. Thus, as expected, one can trade off error for spanner sparsity, increasing
the stretch k to pay more and more error for sparser and sparser spanners.
For very large graphs, purely additive error is arguably a much more appealing paradigm.
For a constant c > 0, a +c spanner of an n-vertex graph G is a subgraph H such that
distH (s, t) ≤ distG (s, t) + c for all vertices s, t. Thus, for additive error the excess distance
in H is independent of the graph size and of distG (s, t), which can be large when n is large.
Additive spanners were introduced by Liestman and Shermer [28], and followed by three
landmark theoretical results on the sparsity of additive spanners in unweighted graphs:
Aingworth, Chekuri, Indyk, and Motwani [7] showed that all graphs have +2 spanners on
O(n3/2 ) edges, Chechik [17, 13] showed that all graphs have +4 spanners on O(n7/5 ) edges,
and Baswana, Kavitha, Mehlhorn, and Pettie [11] showed that all graphs have +6 spanners
on O(n4/3 ) edges.
Despite the inherent appeal of additive error, spanners with multiplicative error remain
much more commonly used in practice. There are two reasons for this.
1. First, while the multiplicative spanner of Althöfer et al [8] works without issue for weighted
graphs, the previous additive spanner constructions hold only for unweighted graphs,
whereas the metrics that arise in applications often require edge weights. Addressing this,
recent work of the authors [3] and in two papers of Elkin, Gitlitz, and Neiman [22, 23]
gave natural extensions of the classic additive spanner constructions to weighted graphs.
For example, the +2 spanner bound becomes the following statement: for all n-vertex
weighted graphs G, there is a subgraph H satisfying distH (s, t) ≤ distG (s, t) + 2W (s, t),
where W (s, t) denotes the maximum edge weight along an arbitrary s t shortest path
in G. The +4 spanner generalizes similarly, and the +6 spanner does as well with the
small exception that the error increases to +(6 + ε)W (s, t), for ε > 0 arbitrarily small
which trades off with the implicit constant in the spanner size.
2. Second, poly(n) factors in spanner size can be quite serious in large graphs, and so
applications often require spanners of near-linear size, say O(n1.01 ) edges for an n-vertex
input graph. The worst-case spanner sizes of O(n4/3 ) or greater for additive spanner
constructions are thus undesirable, and unfortunately, there is a theoretical barrier to
improving them: Abboud and Bodwin [1] proved that one cannot generally trade off more
additive error for sparser spanners, as one can in the multiplicative setting. Specifically,
for any constant c > 0, there is no general construction of +c spanners for n-node
input graphs on O(n4/3−0.001 ) edges. However, the lower bound construction is rather
pathological, and it is not likely to arise in practice. It is known that for many practicalXX:2 Multi-level Weighted Additive Spanners
graph classes, e.g., those with good expansion, near-linear additive spanners always exist
[11]. Thus, towards applications of additive error, it is currently an important open
question whether modern additive spanner constructions on practical graphs of interest
tend to exhibit performance closer to the worst-case bounds from [1], or bounds closer
to the best ones available for the given input graphs. (We remark here that there are
strong computational barriers to designing algorithms that achieve the sparsest possible
+c spanners directly, or which even closely approximate this quantity in general [18]).
The goal of this work is to address the second point, by measuring the experimental
performance of the state-of-the-art constructions for weighted additive spanners on graphs
generated from various popular random models and measuring their performance. We
consider both +cW spanners (where W = maxuv∈E w(uv) is the maximum edge weight)
and +cW (·, ·) spanners, whose additive error is +cW (s, t) for each pair s, t ∈ V . We are
interested both in runtime and in the ratio of output spanner size to the size of the sparsest
possible spanner (which we obtain using an ILP, discussed in Section 5). We specifically
consider generalizations of the three staple constructions for weighted additive spanners
mentioned above, in which the spanner distance constraint only needs to be satisfied for
given pairs of vertices.
In particular, the following extensions are considered. A pairwise spanner is a subgraph
that must satisfy the spanner error inequality for a given set of vertex pairs P taken on input,
and a subsetwise spanner is a pairwise spanner with the additional structure P = S × S for
some vertex subset S. See [31, 21, 27, 26, 12, 20, 14, 15] for recent prior work on pairwise
and subsetwise spanners. We also discuss a multi-level version of the subsetwise additive
spanner problem where we have an edge-weighted graph G = (V, E), a nested sequence
of terminals S` ⊆ S`−1 ⊆ · · · ⊆ S1 ⊆ V and a real number c ≥ 0 as input. We want to
compute a nested sequence of subgraphs G` ⊆ G`−1 ⊆ · · · ⊆ G1 such that Gi is a +cW
subsetwise spanner of G over Si . The objective is to minimize the total number of edges in
all subgraphs. Similar generalizations have been studied for the Steiner tree problem under
various names including Multi-level Network Design [9], Quality of Service Multicast Tree
(QoSMT) [16, 25], Priority Steiner Tree [19], Multi-Tier Tree [29], and Multi-level Steiner
Tree [2, 6]. However, multi-level or QoS generalizations of spanner problems appear to have
been much less studied in literature. Section 2 generalizes the +2 subsetwise construction [21],
and Section 4 generalizes to the multi-level setting.
2 Subsetwise spanners
All unweighted
p graphs have polynomially constructible +2 subsetwise spanners over S ⊆ V
on O(n |S|) edges [31, 21]. For weighted graphs, p Ahmed et al. [3] recently give a +4W
subsetwise spanner construction, also using O(n |S|) edges. In this section we show how
to generalize the +2 subsetwise construction [31, 21] to the weighted setting by giving a
construction which produces a subsetwise
p +2W spanner of a weighted graph (with integer
edge weights in [1, W ]) on O(nW |S|) edges. Due to space, we omit most proof details here
but describe the construction instead.
A clustering C = {C1 , C2 , . . . , Cq } is a set of disjoint subsets of vertices. Initially, every
vertex is unclustered. The subsetwise +2W construction has two steps: the clustering phase
and the path buying phase. The clustering phase is exactly the same p as that of [31, 21] in
which we construct a cluster subgraph GC as follows: set β = logn |S|W , and while there
is a vertex v with at least dnβ e unclustered neighbors, we add a cluster C to C containing
exactly dnβ e unclustered neighbors of v (note that v 6∈ C). We add to GC all edges vxR. Ahmed, et al. XX:3
(x ∈ C) and xy (x, y ∈ C). When there are no more vertices with at least dnβ e unclustered
neighbors, we add all the unclustered vertices and their incident edges to GC .
In the second (path-buying)
|S|
phase, we start with a clustering C and a cluster subgraph
G0 := GC . There are z := 2 unordered pairs of vertices in S; let π1 , π2 , . . . , πz denote
the shortest paths between these vertex pairs where πi = π(ui , vi ) has endpoints {ui , vi }. As
in [21], we iterate from i = 1 to i = z and determine whether to add path πi to the spanner.
Define the cost and value of a path πi as follows:
cost(πi ) := # edges of πi which are absent in Gi−1
value(πi ) := # pairs (x, C) where x ∈ {ui , vi }, C ∈ C,
C contains at least one vertex in πi ,
and distπi (x, C) < distGi−1 (x, C)
If cost(πi ) ≤ (2W + 1)value(πi ), then we add (“buy”) πi to the spanner by letting Gi =
Gi−1 ∪ πi . Otherwise, we do not add πi , and let Gi = Gi−1 . The final spanner returned is
H = Gz .
I Lemma 1. For any ui , vi ∈ S, we have distH (ui , vi ) ≤ distG (ui , vi ) + 2W .
Proof sketch. The proof is largely the same as in [21] except with one main difference: in
the unweighted case, the distance between any two points within the same cluster is at most
2, and it is shown in [21] that using this property, if there are t edges in π(u, v) missing
from GC , then there are at least 2t clusters containing at least one vertex on π(u, v). In the
weighted case, assuming W is constant, there are Ω(t) clusters of C which contain at least
one vertex on π(u, v). J
I Corollary 2. Let G be a weighted graph with integer edge weights in [1, W ]. Then G has a
+6W pairwise spanner on O(W n|P |1/4 ) edges.
This follows from applying the +8W construction of Ahmed et al. [3] (Appendix 3, Al-
gorithm 3), except we use the above +2W subsetwise spanner instead of the +4W subsetwise
spanner construction given in [3] as a subroutine.
3 Pairwise spanner constructions [3]
Here, we provide pseudocode (Algorithms 1–3) describing the +2W , +4W , and +8W pairwise
spanner constructions1 by Ahmed et al. [3]. These spanner constructions have a similar
theme: first, construct a d-light initialization, which is a subgraph H obtained by adding
the d lightest edges incident to each vertex (or all edges if the degree is at most d). Then for
each pair (s, t) ∈ P , consider the number of edges in π(s, t) which are absent in the current
subgraph H. Add π(s, t) to H if the number of missing edges is at most some threshold `,
or otherwise randomly sample vertices and either add a shortest path tree rooted at these
vertices, or construct a subsetwise spanner among them.
1
Using a tighter analysis or the above +2W subsetwise construction in place of the +4W construction in
Algorithm 3, the additive error can be improved to +2W (·, ·), +4W (·, ·), and +6W for integer edge
weights.XX:4 Multi-level Weighted Additive Spanners
Algorithm 1 +2W pairwise spanner [3]
1: d = |P |1/3 , ` = n/|P |2/3
2: H = d-light initialization
3: let m0 be the number of missing edges needed for a valid construction
4: while m0 > nd do
5: for (s, t) ∈ P do
6: x = |E(π(s, t)) \ E(H)|
7: if x ≤ ` then
8: add π(s, t) to H
9: R = random sample of vertices, each with probability 1/(`d)
10: for r ∈ R do
11: add a shortest path tree rooted at r to each vertex
12: add the m0 missing edges
13: return H
Algorithm 2 +4W pairwise spanner [3]
1: d = |P |2/7 , ` = n/|P |5/7
2: H = d-light initialization
3: let m0 be the number of missing edges needed for a valid construction
4: while m0 > nd do
5: for (s, t) ∈ P do
6: x = |E(π(s, t)) \ E(H)|
7: if x ≤ ` then
8: add π(s, t) to H
9: else if x ≥ n/d2 then
10: R1 = random sample of vertices, each w.p. d2 /n
11: add a shortest path tree rooted at each r ∈ R1
12: else
13: add first ` and last ` missing edges of π(s, t) to H
14: R2 = i.i.d. sample of vertices, w.p. 1/(`d)
15: for each r, r0 ∈ R2 do
16: if exists r → r0 path missing ≤ n/d2 edges then
17: add to H a shortest r → r0 path among paths missing ≤ n/d2 edges
18: add the m0 missing edges
19: return H
4 Multi-level spanners
Here we study a multi-level variant of graph spanners. We first define the problem:
I Definition 3 (Multi-level weighted additive spanner). Given a weighted graph G(V, E) with
maximum weight W , a nested sequence of subsets of vertices S` ⊆ S`−1 ⊆ . . . ⊆ S1 ⊆ V , and
c ≥ 0, a multi-level additive spanner is a nested sequence of subgraphs G` ⊆ G`−1 ⊆ . . . ⊆
G1 ⊆ G, where Gi is a subsetwise +cW spanner over Si .
Observe that Definition 3 generalizes the subsetwise spanner, which is a special case
where ` = 1. We measure the size of a multi-level spanner by its sparsity, defined by
P`
sparsity({Gi }`i=1 ) := i=1 |E(Gi )|.R. Ahmed, et al. XX:5
Algorithm 3 +8W pairwise spanner [3]
1: d = |P |1/4 , ` = n/|P |3/4
2: H = d-light initialization
3: let m0 be the number of missing edges needed for a valid construction
4: while m0 > nd do
5: for (s, t) ∈ P do
6: x = |E(π(s, t)) \ E(H)|
7: if x ≤ ` then
8: add π(s, t) to H
9: else
10: add first ` and last ` missing edges of π(s, t) to H
11: R = random sample of vertices, each w.p. 1/(`d)
12: H 0 = +4W subsetwise (R × R)-spanner [3]
13: add H 0 to H
14: add the m0 missing edges
15: return H
The problem can equivalently be phrased in terms of priorities and rates: each vertex
v ∈ S1 has a priority P (v) between 1 and ` (namely, P (v) = max{i : v ∈ Si }), and we wish
to compute a single subgraph containing edges of different rates such that for all u, v ∈ S1 ,
there is a +cW spanner path consisting of edges of rate at least min{P (u), P (v)}. With this,
we will typically refer to the priority of v to denote the highest i such that v ∈ Si , or 0 if
v 6∈ S1 . In this section, we show that the multi-level version is not significantly harder than
the ordinary “single-level” version: a subroutine which can compute an additive spanner
can be used to compute a multi-level spanner whose sparsity is comparably good. Let OPT
denote the minimum sparsity over all candidate multi-level additive spanners.
We first describe a simple rounding-up approach based on an algorithm by Charikar et
al. [16] for the QoSMT problem, a similar generalization of the Steiner tree problem. For this
approach, assume we have a subroutine which computes an exact or approximate single-level
subsetwise spanner. Given v ∈ S1 , let P (v) ∈ [1, `] denote the priority of v. The rounding-up
approach is as follows: for each v, round P (v) up to the nearest power of 2. This effectively
constructs a “rounded-up” instance where all vertices in S1 have priority either 1, 2, 4, . . . ,
2dlog2 `e . The sparsity of the optimum solution in the rounded-up instance is at most 2OPT;
given the optimum solution to the original instance with sparsity OPT, a feasible solution to
the rounded-up instance with sparsity at most 2OPT can be obtained by rounding up the
rate of each edge to the nearest power of 2.
For each i ∈ {1, 2, 4, . . . , 2dlog2 `e }, use the subroutine to compute a level-i subsetwise
spanner over all vertices whose rounded-up priority is at least i. The final multi-level additive
spanner is obtained by taking the union of these computed spanners, by keeping an edge at
the highest level it appears in.
I Theorem 4. Assuming an exact subsetwise spanner subroutine, the solution computed by
the rounding-up approach has sparsity at most 4 · OPT.
This is proved using the same ideas as the 4ρ-approximation for QoSMT [16]. As
mentioned earlier, in practice we use an approximation algorithm to compute the subsetwise
spanner instead of computing the minimum spanner.XX:6 Multi-level Weighted Additive Spanners
(a) (b)
Figure 1 (a) The rounding-up approach computes an optimal spanner at each level (assuming an
exact subroutine), so the sizes of the spanners on each level are at most that of the optimal solution
(9 + 40 edges vs. 12 + 40). (b) However, when an edge is present in a top-level solution, it must be
present in lower-level solutions. The rounding-up approach takes the union of the spanners in the
bottom level; in this case, the sparsity of the rounded-up solution (9 + 48 vs. 12 + 40) is greater
than that of the optimum.
p
I Theorem 5. There exists a Õ(n/ |S1 |)-approximation algorithm to compute multi-level
weighted additive spanners with additive stretch 2W when W = O(log n).
This follows from using the +2W subsetwise constructionp in Section 2. The approximation
ratio of this subsetwise spanner
p algorithm is O(nW/ |S|) as the construction produces
a spanner of size O(nW |S|), while the sparsest additive spanner trivially has at least
|S| − 1 = Ω(|S|) edges.
We now show that, under certain conditions, if we have a subroutine which computes a
subsetwise spanner of G, S of size O(na |S|b ) where a and b are absolute constants, a very
naïve algorithm can be used to obtain a multi-level spanner also with sparsity O(na |S1 |b ).
I Theorem 6. Suppose there is an absolute constant 0 < α < 1 such that |Si | ≤ α|Si−1 | for
all i ∈ {1, . . . , `}. Then we can compute a multi-level spanner with sparsity O(na |S1 |b ).
Proof. Consider the following simple construction: for each i ∈ {1, 2, 3, . . . , `}, compute
a level-i subsetwise spanner of size O(na |Si |b ). Consider the union of these spanners, by
keeping each edge at the highest level it appears. The sparsity of the returned multi-level
spanner is at most
sparsity({Gi }) = O(na |S1 |b + 2na |S2 |b + 3na |S3 |b + . . . + `na |S` |b )
≤ O(na |S1 |b (1 + 2αb + 3α2b + . . . + `α(`−1)b ))
= O(na |S1 |b )
1
where we used the arithmetico-geometric series 1 + 2(αb ) + 3(αb )2 + . . . = (1−α b )2 which is
b
constant for fixed α, b. Note that 0 < α < 1 and b > 0, which implies 0 < α < 1. J
The assumption that |Si | ≤ α|Si−1 | for some constant α is fairly natural, as many realistic
networks tend to have significantly fewer hubs than non-hubs.
I Corollary 7. Under the assumption |Si | ≤ α|Si−1 | for all i ∈ {2, . . . , `}, p
there exists a
poly-time algorithm which computes a multi-level +2 spanner of sparsity O(n |S1 |).R. Ahmed, et al. XX:7
p
Proof. This follows by using the +2 construction by Cygan et al. [21] on O(n |S|) edges as
the subroutine. J
5 Exact algorithm
To compute a minimum size additive spanner, we utilize a slight modification of the ILP in [5,
Section 9], wherein we choose the specific distortion function f (t) = t + cW and minimize the
sparsity rather than total weight of the spanner. For completeness, we present the full ILP
for computing a single-level additive subsetwise spanner below along with a brief description
of the multi-level extension. Here E 0 represents the bidirected edge set, obtained by adding
directed edges (u, v) and (v, u) for each edge uv ∈ E. The binary variable xuv (i,j) is 1 if edge
(i, j) is included on the selected u-v path and 0 otherwise, and w(e) is the weight of edge e.
X
Minimize xe subject to (1)
e∈E
X
xuv
(i,j) w(e) ≤ distG (u, v) + cW ∀(u, v) ∈ S; e = ij (2)
(i,j)∈E 0
1
i=u
X X
xuv
(i,j) − xuv
(j,i) = −1 i=v ∀(u, v) ∈ S; ∀i ∈ V (3)
(i,j)∈Out(i) (j,i)∈In(i)
0 else
X
xuv
(i,j) ≤ 1 ∀(u, v) ∈ S; ∀i ∈ V (4)
(i,j)∈Out(i)
xuv uv
(i,j) + x(j,i) ≤ xe ∀(u, v) ∈ S; ∀e = {i, j} ∈ E (5)
xe , xuv
(i,j) ∈ {0, 1} (6)
Inequalities (3)–(4) enforce that for each u, v ∈ S, the selected edges corresponding to u,
v form a path; inequality (2) enforces that the length of this path is at most distG (u, v) + cW
(note that W may be replaced with W (u, v)). Inequality (5) ensures that if xuv (i,j) = 1 or
xuv
(i,j) = 1, then edge ij is taken.
To generalize the ILP formulation to the multi-level problem, we take a similar set of
variables for every level. The rest of the constraints are similar, except we define xke = 1 if
edge e is present on level k and the variables xuv (i,j) are also indexed by level. We add the
k k−1
constraint xe ≤ xe for all k ∈ {2, . . . , `} which enforces that if edge e is present on level
k, it is also present on all lower levels. Finally, the objective is to minimize the sparsity
P` P k
k=1 e∈E xe .
6 Experiments
In this section, we provide experimental results involving the rounding-up framework described
in Section 4. This framework needs a single level subroutine; we use the +2W subsetwise
construction in Section 2 and the three pairwise +2W (·, ·), +4W (·, ·), +6W constructions
provided in [3]2 (see Appendix 3). We generate multi-level instances and solve the instances
2
Note that, one can show that the +2W , +4W , +8W spanners in [3] are actually +2W (., .), +4W (., .)
and +6W spanners respectively by using a tighter analysis [4].XX:8 Multi-level Weighted Additive Spanners
using our exact algorithm and the four approximation algorithms. We consider natural
questions about how the number of levels `, number of vertices n, and decay rate of terminals
with respect to levels affect the running times and (experimental) approximation ratios,
defined as the sparsity of the returned multi-level spanner divided by OPT.
We used CPLEX 12.6.2 as an ILP solver in a high-performance computer for all exper-
iments (Lenovo NeXtScale nx360 M5 system with 400 nodes). Each node has 192 GB of
memory. We have used Python for implementing the algorithms and spanner constructions.
Since we have run the experiment on a couple of thousand instances, we run the solver for
four hours.
6.1 Experiment Parameters
We run experiments first to test experimental approximation ratio vs. the parameters, and
then to test running time vs. parameters. Each set of experiments has several parameters:
the graph generator, the number of levels `, the number of vertices n, and how the size of
the terminal sets Si (vertices requiring level or priority at least i) decrease as i decreases.
In what follows, we use the Erdős–Rényi (ER) [24], Watts–Strogatz (WS) [32], Barabási–
Albert (BA) [10], and random geometric (GE) [30] models. Let p be the edge selection
probability. If we set p = (1 + ε) lnnn , then the generated Erdős–Rényi graph is connected with
high probability for ε > 0 [24]). For our experiments we use ε = 1. In the Watts-Strogatz
model, we initially create a ring lattice of constant degree K. For our experiments we use
K = 6 and p = 0.2. In the Barabási–Albert model, a new node is connected to m existing
nodes. For our experiments we use m = 5. In the random geometric graph model, two nodes
are connected
q to each other if their Euclidean distance is not larger than a threshold rc . For
ln n
rc = (1+) πn with > 0, the synthesized graph is connected with a high probability[30].
We generate a set of small graphs (10 ≤ n ≤ 40) and a set of large graphs (50 ≤ n ≤ 500).
We only compute the exact solutions for the small graphs since the ILP has an exponential
running time. In this paper, we provide the results of Erdős–Rényi graphs since it is the
most popular model. However, the radius3 of Erdős–Rényi graphs is relatively small. In our
dataset, the range of the radius is 2-4. Hence, we also provide the results of random geometric
graphs which have larger radius (4-12). The remaining results and the radius distribution
of different generators are available at the supplement Github link. We consider number of
levels ` ∈ {1, 2, 3} for small graphs, ` ∈ {1, . . . , 10} for large graphs, and adopt two methods
for selecting terminal sets: linear and exponential. A terminal set S1 with lowest priority
of size n(1 − `+1 1
) in the linear case and n2 in the exponential case is chosen uniformly at
1
random. For each subsequent level, `+1 vertices are deleted at random in the linear case,
whereas half the remaining vertices are deleted in the exponential case. Levels/priorities
and terminal sets are related via Si = {v ∈ S1 : P (v) ≥ i}. We choose edge weights w(e)
randomly, independently, and uniformly from {1, 2, 3 . . . , 10}.
An experimental instance of the multi-level problem here is thus characterized by four
parameters: graph generator, number of vertices n, number of levels `, and terminal selection
method TSM ∈ {Linear,Exponential}. As there is randomness involved, we generated
five instances for every choice of parameters (e.g., ER, n = 30, ` = 2, Linear). For each
instance of the small graphs, we compute the approximate solution using either the +2W ,
+2W (·, ·), +4W , or +6W spanner subroutine, and the exact solution using the ILP described
3
The minimum over all v ∈ V of maxw∈V dG (v, w) where dG (v, w) is the graph distance (by number of
edges, not total weight) between v and wR. Ahmed, et al. XX:9
in Section 5. We compute the experimental approximation ratio by dividing the sparsity of
the approximate solution by the sparsity of the optimum solution (OPT). For large graphs,
we only compute the approximate solution.
6.2 Results
We consider different spanner constructions as the single level subroutine in the multi-level
spanner. We first consider the +2W subsetwise construction (Section 2).
6.2.1 The +2W Subsetwise Construction-based Approximation
We first describe the experimental results on Erdős–Rényi graphs w.r.t. n, `, and terminal
selection method in Figure 2. The average experimental ratio increases
p as n increases. This
is expected since the theoretical approximation ratio of Õ(n/ |S1 |) is proportional to n.
The average and minimum experimental ratio does not change that much as the number of
levels increases; however, the maximum ratio increases. The experimental ratio of the linear
terminal selection method is slightly better compared to that of the exponential method.
Figure 2 Performance of the algorithm that uses +2W subsetwise spanner as the single level
solver on Erdős–Rényi graphs w.r.t. n, `, and terminal selection method.
We describe the experimental results on random geometric graphs w.r.t. n, `, and terminal
selection methods in Figure 3. In both cases the average ratio increases as n and ` increases.
The average ratio is relatively lower for the linear terminal selection method.
Figure 3 Performance of the algorithm that uses +2W subsetwise spanner as the single level
solver on random geometric graphs w.r.t. n, `, and terminal selection method.
The plots of Watts–Strogatz and Barabási–Albert graphs are available in the Github
repository.
6.2.2 The +2W (·, ·) Pairwise Construction-based Approximation
We now consider the +2W (·, ·) pairwise construction [3] (Algorithm 1). We first describe the
experimental results on Erdős–Rényi graphs w.r.t. n, `, and terminal selection method inXX:10 Multi-level Weighted Additive Spanners
Figure 4. The average experimental ratio increases as n increases. This is expected since the
theoretical approximation ratio is proportional to n. The average and minimum experimental
ratio do not change that much as the number of levels increases, however, the maximum
ratio increases. The experimental ratio of the linear terminal selection method is also slightly
better compared to that of the exponential method.
Figure 4 Performance of the algorithm that uses +2W (·, ·) pairwise spanner as the single level
solver on Erdős–Rényi graphs w.r.t. n, `, and terminal selection method.
We describe the experimental results on random geometric graphs w.r.t. n, `, and terminal
selection method in Figure 5. The average experimental ratio increases as n increases. The
maximum ratio increases as ` increases. Again, the experimental ratio of the linear terminal
selection method is relatively smaller compared to the exponential method.
Figure 5 Performance of the algorithm that uses +2W (·, ·) pairwise spanner as the single level
solver on random geometric graphs w.r.t. n, `, and terminal selection method.
6.2.3 Comparison between Global and Local Setups
One major difference between the subsetwise and pairwise construction is the subsetwise
construction considers the (global) maximum edge weight W of the graph in the error. On the
other hand, the +cW (·, ·) spanners consider the (local) maximum edge weight in a shortest
path for each pair of vertices s, t. We provide a comparison between the global and local
settings.
We describe the experimental results on Erdős–Rényi graphs w.r.t. n, `, and the terminal
selection method in Figure 6. The average experimental ratio increases as n increases for
both global and local settings. However, the ratio of the local setting is smaller compared to
that of the global setting. One reason for this difference is the solution to the global exact
algorithm is relatively smaller since the global setting considers larger errors. The ratio of
the global setting increases as the number of levels increases and for the exponential terminal
selection method. For the local setting, the ratio does not change that much.
We describe the experimental results on random geometric graphs w.r.t. n, `, and the
terminal selection method in Figure 7. The ratio of the local setting is smaller compared toR. Ahmed, et al. XX:11
Figure 6 Performance of the global and local construction-based algorithms on Erdős–Rényi
graphs w.r.t. n, `, and terminal selection method.
the global setting.
Figure 7 Performance of the global and local construction-based algorithms on random geometric
graphs w.r.t. n, `, and terminal selection method.
6.2.4 The +4W (·, ·) Pairwise Construction-based Approximation
We now consider the +4W (·, ·) pairwise construction [3] (Algorithm 2) as a single level
subroutine. We first describe the experimental results on Erdős–Rényi graphs w.r.t. n, `,
and terminal selection method in Figure 8. The average experimental ratio increases as n
increases. This is expected since the theoretical approximation ratio is proportional to n.
The average experimental ratio does not change that much as the number of levels increases;
however, the maximum ratio increases. The experimental ratio of the linear terminal selection
method is also slightly better compared to that of the exponential method.
Figure 8 Performance of the algorithm that uses +4W (·, ·) pairwise spanner as the single level
solver on Erdős–Rényi graphs w.r.t. n, `, and terminal selection method.
We describe the experimental results on random geometric graphs w.r.t. n, `, and
terminal selection method in Figure 9. The experimental ratio increases as the number of
vertices increases. The maximum ratio increases as the number of levels increases. Again,
the experimental ratio of the linear terminal selection method is relatively smaller compared
to the exponential method.XX:12 Multi-level Weighted Additive Spanners
Figure 9 Performance of the algorithm that uses +4W (·, ·) pairwise spanner as the single level
solver on random geometric graphs w.r.t. n, `, and terminal selection method.
6.2.5 Comparison between +2W (·, ·) and +4W (·, ·) Setups
We now provide a comparison between the pairwise +2W (·, ·) and +4W (·, ·) construction-
based approximation algorithms. We first describe the experimental results on Erdős–Rényi
graphs w.r.t. n, `, and the terminal selection method in Figure 10. The average experimental
ratio increases as n increases for both +2W (·, ·) and +4W (·, ·) settings. As n increases,
the ratio of the +4W (·, ·) construction-based algorithm decreases slightly. The +4W (·, ·)
construction-based algorithm slightly outperforms the +2W (·, ·) algorithm for ` = 3 and
exponential selection method.
Figure 10 Performance of the pairwise +2W (·, ·) and +4W (·, ·) construction-based algorithms
on Erdős–Rényi graphs w.r.t. n, `, and terminal selection method.
We describe the experimental results on random geometric graphs w.r.t. n, `, and the
terminal selection method in Figure 11. As n increases the average ratio of +4W (·, ·)-based
approximation algorithm becomes smaller compared to the +2W (·, ·)-based algorithm. The
average ratio of +4W (·, ·) is relatively smaller for the exponential terminal selection method.
Figure 11 Performance of the pairwise +2W (·, ·) and +4W (·, ·) construction-based algorithms
on random geometric graphs w.r.t. n, `, and terminal selection method.R. Ahmed, et al. XX:13
6.2.6 The +6W Pairwise Construction-based Approximation
We now consider the +6W pairwise construction [3] (Algorithm 3) as a single level solver.
We first describe the experimental results on Erdős–Rényi graphs w.r.t. n, `, and terminal
selection method in Figure 12. The average experimental ratio increases as n increases.
This is expected since the theoretical approximation ratio is proportional to n. The average
experimental ratio does not change that much as the number of levels increases; however,
the maximum ratio increases. The maximum and average experimental ratios of the linear
terminal selection method are slightly better compared to that of the exponential method.
Figure 12 Performance of the algorithm that uses +6W pairwise spanner as the single level
solver on Erdős–Rényi graphs w.r.t. n, `, and terminal selection method.
We describe the experimental results on random geometric graphs w.r.t. n, `, and
terminal selection method in Figure 13. The experimental ratio increases as the number of
vertices increases. The maximum ratio increases as the number of levels increases. Again,
the experimental ratio of the linear terminal selection method is relatively smaller compared
to the exponential method.
Figure 13 Performance of the algorithm that uses +6W (·, ·) pairwise spanner as the single level
solver on random geometric graphs w.r.t. n, `, and terminal selection method.
6.2.7 Comparison between +2W and +6W Setups
We now provide a comparison between pairwise +2W and +6W construction-based ap-
proximation algorithms. We first describe the experimental results on Erdős–Rényi graphs
w.r.t. n, `, and the terminal selection method in Figure 14. The average experimental
ratio increases as n increases for both +2W and +6W settings. As the number of vertices
increases, the ratio of the +6W construction-based algorithm gets smaller. This is expected
since a larger error makes the problem easier to solve. Similarly, as ` increases, the +6W
construction-based algorithm outperforms the +2W algorithm. The average experimental
ratio of the +6W construction based algorithm is smaller both in the linear and exponential
terminal selection methods.XX:14 Multi-level Weighted Additive Spanners
Figure 14 Performance of the pairwise +2W and +6W construction-based algorithms on Erdős–
Rényi graphs w.r.t. n, `, and terminal selection method.
We describe the experimental results on random geometric graphs w.r.t. n, `, and the
terminal selection method in Figure 15. We can see that as n gets larger the ratio of +6W
gets smaller. The situation is similar when ` increases.
Figure 15 Performance of the pairwise +2W and +6W construction-based algorithms on random
geometric graphs w.r.t. n, `, and terminal selection method.
6.2.8 Experiment on Large Graphs
We generate some large instances on up to 500 vertices and run different multi-level spanner
algorithms on them. We use n = {50, 100, 150, . . . , 500} and ` = {1, 2, 3, . . . , 10}. We describe
the experimental results on Erdős–Rényi graphs w.r.t. n, `, and the terminal selection method
in Figure 16. We are comparing four multi-level algorithms, namely those using the +2W
subsetwise and +2W (·, ·), +4W (·, ·), +6W pairwise constructions [3] as subroutines with
P = S × S. Since computing the optimal solution exactly via ILP is computationally
expensive on large instances, we report the ratio in terms of relative sparsity, defined as
the sparsity of the multi-level spanner returned by one algorithm divided by the minimum
sparsity over the spanners returned by all four. The ratio of the +6W construction based
algorithm is lowest and the +2W construction based algorithm is highest. This is expected
since a higher additive error generally reduces the number of edges needed. Overall the ratio
decreases as n increases. This is because the significance of small additive error reduces as
the graph size and distances get larger. The relative ratio for the +2W construction increases
as ` increases, and for the exponential terminal selection method.
We describe the experimental results on random geometric graphs w.r.t. n, `, and the
terminal selection method in Figure 17.R. Ahmed, et al. XX:15
Figure 16 Performance of different approximation algorithms on large Erdős–Rényi graphs w.r.t.
n, `, and terminal selection method.
Figure 17 Performance of different approximation algorithms on large random geometric graphs
w.r.t. n, `, and terminal selection method.
6.2.9 Impact of Initial Parameters
It is worth mentioning that the +2 subsetwise spanner [21] and +2W subsetwise spanner
(Section 2) begin with a clustering phase, while the algorithms described in Appendix 3 begin
with a d-light initialization. In d-light initialization, we add the d lightest edges incident to
each vertex; these edges tend to be on shortest paths. In practice, there may be relatively few
edges which appear on shortest paths and some of these edges might be redundant. Hence,
we compute +2W (·, ·) spanners with different values of d. We describe the experimental
results on Erdős–Rényi graphs w.r.t. n, `, and the terminal selection method in Figure 18.
We have computed the ratio as described in Section 6.2.8. From the figures, we see that
as we reduce the value of d exponentially, the ratio decreases: in particular, the optimal
choice of parameter d in practice might be significantly smaller than the optimal value of d
in theory. Generally, it could make sense in practical implementations of spanner algorithms
to try all values {d, d/2, d/4, d/8, . . . }, computing log d different spanners, and then use only
the sparsest one. This costs only a log d factor in the running time of the algorithm, which
is typically reasonable.
Figure 18 Impact of different values of d on large Erdős–Rényi graphs w.r.t. n, `, and terminal
selection method.XX:16 Multi-level Weighted Additive Spanners
We describe the experimental results on random geometric graphs w.r.t. n, `, and
the terminal selection method in Figure 19. Again, the experiment suggests that we can
exponentially reduce the value of d and take the solution that has a minimum number of
edges.
Figure 19 Impact of different values of d on large random geometric graphs w.r.t. n, `, and
terminal selection method.
6.3 Running Time
We now provide the running times of the different algorithms. We show the running time of
the ILP on Erdős–Rényi graphs w.r.t. n, `, and terminal selection method in Figure 20. The
running time of the ILP increases exponentially as n increases, as expected. The execution
time of a single level instance with 45 vertices is more than 64 hours using a 28 core processor.
Hence, we kept the number of vertices less than or equal to 40 for our small graphs. The
experimental running time should increase as ` increases, but we do not see that pattern in
these plots because some of the instances were not able to finish in four hours.
Figure 20 Running time of all exact algorithms on Erdős–Rényi graphs w.r.t. n, `, and terminal
selection method.
We provide the experimental running time of the approximation algorithm on Erdős–
Rényi graphs in Figure 21. The running time of the +2W construction-based algorithm is
the largest. Overall, the running time increases as n increases. There is no straightforward
relation between the running time and `. Although the number of calls to the single level
subroutine increases as ` increases, it also depends on the size of the subset in a single level.
Hence, if the subset sizes are larger, then it may take longer for small `. The running time of
the linear method is larger.
The running times appear reasonable in other settings too; see the supplemental Github
repository for details and experimental results.
7 Conclusion
We have provided a framework where we can use different spanner subroutines to compute
multi-level spanners of varying additive error. We additionally introduced a generalization ofR. Ahmed, et al. XX:17
Figure 21 Running time of all approximation algorithms on large Erdős–Rényi graphs w.r.t. n, `,
and terminal selection method.
the +2 subsetwise spanner [21] to integer edge weights, and illustrate that this can reduce
the +8W error in [3] to +6W . A natural question is to provide an approximation algorithm
that can handle different additive error for different levels. We also provided an ILP to find
the exact solution for both global and local settings. Using the ILP and the given spanner
constructions, we have run experiments and provided the experimental approximation ratio
over the graphs we generated. The experimental results suggest that the +2W clustering-
based approach is slower than the initialization based approaches. We provided a method of
changing the initialization parameter d which reduces the sparsity in practice.
References
1 Amir Abboud and Greg Bodwin. The 4/3 additive spanner exponent is tight. Journal of the
ACM (JACM), 64(4):1–20, 2017.
2 Abu Reyan Ahmed, Patrizio Angelini, Faryad Darabi Sahneh, Alon Efrat, David Glickenstein,
Martin Gronemann, Niklas Heinsohn, Stephen Kobourov, Richard Spence, Joseph Watkins, and
Alexander Wolff. Multi-level Steiner trees. In 17th International Symposium on Experimental
Algorithms, (SEA), pages 15:1–15:14, 2018. URL: https://doi.org/10.4230/LIPIcs.SEA.
2018.15, doi:10.4230/LIPIcs.SEA.2018.15.
3 Reyan Ahmed, Greg Bodwin, Faryad Darabi Sahneh, Stephen Kobourov, and Richard Spence.
Weighted additive spanners. In Isolde Adler and Haiko Müller, editors, Graph-Theoretic
Concepts in Computer Science, pages 401–413, Cham, 2020. Springer International Publishing.
4 Reyan Ahmed, Greg Bodwin, Keaton Hamm, Stephen Kobourov, and Richard Spence.
Weighted sparse and lightweight spanners with local additive error. arXiv preprint
arXiv:2103.09731, 2021.
5 Reyan Ahmed, Greg Bodwin, Faryad Darabi Sahneh, Keaton Hamm, Mohammad Javad Latifi
Jebelli, Stephen Kobourov, and Richard Spence. Graph spanners: A tutorial review. Computer
Science Review, 37:100253, 2020.
6 Reyan Ahmed, Faryad Darabi Sahneh, Keaton Hamm, Stephen Kobourov, and Richard
Spence. Kruskal-Based Approximation Algorithm for the Multi-Level Steiner Tree Problem.
In Fabrizio Grandoni, Grzegorz Herman, and Peter Sanders, editors, 28th Annual European
Symposium on Algorithms (ESA 2020), volume 173 of Leibniz International Proceedings in
Informatics (LIPIcs), pages 4:1–4:21, Dagstuhl, Germany, 2020. Schloss Dagstuhl–Leibniz-
Zentrum für Informatik. URL: https://drops.dagstuhl.de/opus/volltexte/2020/12870,
doi:10.4230/LIPIcs.ESA.2020.4.
7 Donald Aingworth, Chandra Chekuri, Piotr Indyk, and Rajeev Motwani. Fast estimation of
diameter and shortest paths (without matrix multiplication). SIAM Journal on Computing,
28:1167–1181, 04 1999. doi:10.1137/S0097539796303421.
8 Ingo Althöfer, Gautam Das, David Dobkin, and Deborah Joseph. Generating sparse spanners
for weighted graphs. In Scandinavian Workshop on Algorithm Theory (SWAT), pages 26–37,
Berlin, Heidelberg, 1990. Springer Berlin Heidelberg.XX:18 Multi-level Weighted Additive Spanners
9 Anantaram Balakrishnan, Thomas L. Magnanti, and Prakash Mirchandani. Modeling and
heuristic worst-case performance analysis of the two-level network design problem. Management
Sci., 40(7):846–867, 1994. doi:10.1287/mnsc.40.7.846.
10 Albert-László Barabási and Réka Albert. Emergence of scaling in random networks. science,
286(5439):509–512, 1999.
11 Surender Baswana, Telikepalli Kavitha, Kurt Mehlhorn, and Seth Pettie. Additive spanners
and (α, β)-spanners. ACM Transactions on Algorithms (TALG), 7(1):5, 2010.
12 Greg Bodwin. Linear size distance preservers. In Proceedings of the Twenty-Eighth Annual
ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 600–615. Society for Industrial
and Applied Mathematics, 2017.
13 Greg Bodwin. A note on distance-preserving graph sparsification. arXiv preprint
arXiv:2001.07741, 2020.
14 Greg Bodwin and Virginia Vassilevska Williams. Better distance preservers and additive
spanners. In Proceedings of the Twenty-seventh Annual ACM-SIAM Symposium on Discrete
Algorithms (SODA), pages 855–872. Society for Industrial and Applied Mathematics, 2016.
URL: http://dl.acm.org/citation.cfm?id=2884435.2884496.
15 Hsien-Chih Chang, Pawel Gawrychowski, Shay Mozes, and Oren Weimann. Near-Optimal
Distance Emulator for Planar Graphs. In Proceedings of 26th Annual European Symposium
on Algorithms (ESA 2018), volume 112, pages 16:1–16:17, 2018.
16 M. Charikar, J. Naor, and B. Schieber. Resource optimization in QoS multicast routing of
real-time multimedia. IEEE/ACM Transactions on Networking, 12(2):340–348, April 2004.
doi:10.1109/TNET.2004.826288.
17 Shiri Chechik. New additive spanners. In Proceedings of the twenty-fourth annual ACM-SIAM
symposium on Discrete algorithms (SODA), pages 498–512. Society for Industrial and Applied
Mathematics, 2013.
18 Eden Chlamtáč, Michael Dinitz, Guy Kortsarz, and Bundit Laekhanukit. Approximating span-
ners and directed steiner forest: Upper and lower bounds. In Proceedings of the Twenty-Eighth
Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 534–553. SIAM, 2017.
19 Julia Chuzhoy, Anupam Gupta, Joseph (Seffi) Naor, and Amitabh Sinha. On the approx-
imability of some network design problems. ACM Trans. Algorithms, 4(2):23:1–23:17, 2008.
doi:10.1145/1361192.1361200.
20 Don Coppersmith and Michael Elkin. Sparse sourcewise and pairwise distance preservers.
SIAM Journal on Discrete Mathematics, 20(2):463–501, 2006.
21 Marek Cygan, Fabrizio Grandoni, and Telikepalli Kavitha. On pairwise spanners. In
Proceedings of 30th International Symposium on Theoretical Aspects of Computer Science
(STACS 2013), volume 20, pages 209–220, 2013. URL: http://drops.dagstuhl.de/opus/
volltexte/2013/3935, doi:10.4230/LIPIcs.STACS.2013.209.
22 Michael Elkin, Yuval Gitlitz, and Ofer Neiman. Almost shortest paths and PRAM distance
oracles in weighted graphs. arXiv preprint arXiv:1907.11422, 2019.
23 Michael Elkin, Yuval Gitlitz, and Ofer Neiman. Improved weighted additive spanners. arXiv
preprint arXiv:2008.09877, 2020.
24 Paul Erdős and Alfréd Rényi. On random graphs, i. Publicationes Mathematicae (Debrecen),
6:290–297, 1959.
25 Marek Karpinski, Ion I. Mandoiu, Alexander Olshevsky, and Alexander Zelikovsky. Improved
approximation algorithms for the quality of service multicast tree problem. Algorithmica,
42(2):109–120, 2005. doi:10.1007/s00453-004-1133-y.
26 Telikepalli Kavitha. New pairwise spanners. Theory of Computing Systems, 61(4):1011–
1036, Nov 2017. URL: https://doi.org/10.1007/s00224-016-9736-7, doi:10.1007/
s00224-016-9736-7.
27 Telikepalli Kavitha and Nithin M. Varma. Small stretch pairwise spanners and ap-
proximate d-preservers. SIAM Journal on Discrete Mathematics, 29(4):2239–2254, 2015.R. Ahmed, et al. XX:19
URL: https://doi.org/10.1137/140953927, arXiv:https://doi.org/10.1137/140953927,
doi:10.1137/140953927.
28 Arthur Liestman and Thomas Shermer. Additive graph spanners. Networks, 23:343 – 363, 07
1993. doi:10.1002/net.3230230417.
29 Prakash Mirchandani. The multi-tier tree problem. INFORMS J. Comput., 8(3):202–218,
1996.
30 Mathew Penrose. Random geometric graphs. Number 5. Oxford university press, 2003.
31 Seth Pettie. Low distortion spanners. ACM Transactions on Algorithms (TALG), 6(1):7,
2009.
32 Duncan J Watts and Steven H Strogatz. Collective dynamics of ‘small-world’networks. Nature,
393(6684):440, 1998.You can also read
